Creating composite variables

Question

I have a question merging two variables into a single composite. The questions are about marijuana use. One variable is frequency of use measured by the question: "On how many days during the last 30 days did you use marijuana?[0-30]" Quantity is measured with the question: "On a typical marijuana use day in the past 30 days, please indicate how high you get from using marijuana? [0=not at all, 100=completely]". If possible, I want to merge the two scores into one variable representative of someone's marijuana use.

Is there a correct way of doing this? I don't think quantity in itself means a lot.

Answer 1

Proper measurement of consumption of drugs is a very difficult thing. The two questions that you have are not really adequate (although it is quite hard to come up with a set of questions that works).

For your problem of "how to combine" the questions, I would recommend an ad-hoc approach.Look at all the patterns that people in your data set exhibit. Then try to come up with new levels that make sense, given your data. So, you might have, e.g. "Frequent moderate users" and so on.

One problem is that there's no way for a person to say that, e.g., he uses a little pot on 5 days a week and a lot on 2 days a week.

Answer 2

I would advocate against doing what you propose. The frequency question is a completely objective question. How high one gets from use is completely subjective. Further, this tells you nothing of the (objective or subjective) quantity used. Inferring that someone who gets very high uses a lot of pot is incorrect and misleading. You must design a different question to assess quantity used, such as frequency and quantity of purchasing, number of puffs taken or joints/bowls smoked during a session. For some inspiration you can consult other drinking/smoking questionnaires like the Youth Tobacco Survey, or the AUDIT-C.

As it stands, my questionnaire-design perspective suggests that merging these questions into a single coded variable will produce confusing and difficult to interpret output. From an analysis perspective they should be analyzed separately.